dnnyyq / mmdetection-supervisely-person-datasets Goto Github PK

We used mmdetection(mask rcnn) to train supervisely person dataset.

• Ubuntu 16.04
• Python 3.4+
• PyTorch 0.4.1
• Cython

git clone https://github.com/nicehuster/mmdetection-supervisely-person-datasets.git

cd mmdetection-supervisely-person-datasets

bash compile.sh

# run setup
python setup.py install 

# or install locally
python setup.py install --user

export PYTHONPATH=$PYTHONPATH:{pwd}

You can download supervisely person dataset from the official website, or you can download from baiduyun(链接: https://pan.baidu.com/s/1EsXXlEvA_tAkLv1vqgPEKA 提取码: xrm9) after download the dataset, place it in data folder, and The dataset is structured as follows：

.
├── ds1
│   ├── ann
│   └── img
...
├── ds11
│   ├── ann
│   └── img
├── ds12
│   ├── ann
│   └── img
├── ds13
│   ├── ann
│   └── img
└── meta.json

You can download the pretrained model trained on supervisely person dataset in baidu,passwd:ytiv

python demox.py

python tools/train.py configs/supervisely.py --gpus 2 

Important:

• you need to set the learning rate proportional to the GPU num. E.g., modify lr to 0.01 for 4 GPUs or 0.005 for 2 GPUs.
• In this repository, the pytorch version is 0.4.1
• In the pretrained model, I only trained the model 80 epoches, you can train more epoch to get better results.